$518 Billion on Clouds and Chips: Claude’s Makers Reveal the Real Cost of the AI Race Ahead of a $2 Trillion IPO

The developer of the Claude family of neural networks has set aside up to $518 billion for cloud computing leases and infrastructure development in the coming years. The figure appears in a confidential draft prospectus for an initial public offering (IPO) reviewed by Reuters. For the tech sector, the sum looks unprecedented, but it does not represent one-time capital spending or the construction of its own data centers — it is the combined value of long-term, multiyear contracts with third-party capacity providers.

The half-trillion-dollar figure reflects the harsh reality of the AI race: training and maintaining models on the scale of Claude requires guaranteed reservation of power and server resources a decade ahead.

Megawatts from Amazon, Google and a SpaceX Contract

The astronomical total of obligations comes from a series of large-scale partnership agreements the company has signed over the past two years. Its biggest partner is Amazon: the two sides signed a 10-year contract worth more than $100 billion that guarantees Anthropic access to AWS data center power capacity totaling up to 5 GW, according to the company’s announcement.

Similar 5 GW agreements have been signed with the Google-Broadcom alliance, another $30 billion is reserved for Microsoft Azure infrastructure, and a joint project with operator Fluidstack is valued at $50 billion. The partner list even includes aerospace company SpaceX, which has provided more than 300 MW of computing capacity under a direct agreement with the developers.

Just days ago, network corporation Akamai joined the pool of suppliers: Anthropic has committed to paying $11.6 billion over seven years, with the deal potentially expanding to nearly $20 billion, according to Akamai’s official press release.

Revenue Up 12-Fold and Non-Cash Losses

The company’s financial report shows both explosive revenue growth and the enormous cost of maintaining leadership. In 2025, Anthropic’s revenue grew roughly 12-fold, reaching $4.6 billion. At the same time, net spending on computing capacity and clouds alone amounted to $7.33 billion, while the operating loss exceeded $8 billion.

The formal net loss for the reporting period came to almost $42 billion, but finance experts caution against confusing that metric with a cash shortfall. About $34 billion of that sum comes from a non-cash accounting revaluation of convertible securities obligations after previous funding rounds. The company’s actual liquidity cushion remains high: at the end of 2025, the startup held $20.28 billion in cash and short-term investments.

Investment bankers are targeting a valuation of more than $2 trillion when the company goes public. The official draft of Form S-1 has already been submitted confidentially to the SEC, although filing the documents does not legally obligate the company to list on a strictly fixed date.

Notably, the derivatives market took the news of half-trillion-dollar obligations without panic: synthetic perpetual contracts traded on crypto exchanges and tied to Anthropic’s future market capitalization fell just 2%, holding the $2 trillion mark with open interest above $100 million, CoinDesk notes. For the market, the data confirms a simple truth: victory in the AI industry will go to those who can contract the most of the world’s energy and microchips.